运动目标检测方法及其实时实现技术研究

运动目标检测是计算机视觉研究领域的一项重要课题，具有广阔的应用前景和较高的学术价值。实现可靠、快速的运动目标检测并将其相应算法固化为硬件结构具有非常重要的意义。 本文首先深入研究了基于动态场景的运动目标检测算法，详细讨论了相邻帧差法及其改进方法：三帧差分法，并提出了一种融合相位一致性边缘检测的帧间差分方法；接着在研究了背景差分法的基础上讨论了背景获取和更新机制；然后探讨了块匹配法的基本原理及其匹配准则和快速搜索策略。 其次在研究了基于时空梯度光流计算方法的基础上，结合基于全局和局部约束的混合模型，将PM非线性扩散模型引入到了光流计算过程，同时为了加速计算和克服亮度不变模型对大运动矢量的不适应性，采取了多尺度的计算策略。实验结果表明，该方法对噪声具有较强的鲁棒性并且能够较好的保持目标的运动边缘。 最后开发了一种基于Altera公司Cyclone II系列FPGA芯片的运动目标检测系统，详细介绍了系统的硬件电路设计和相关模块的FPGA实现过程。采用FPGA的实现方案，使得系统的设计灵活、集成度高。

VGIS-COLLIDE: an effective collision detection algorithm for

Collision detection is an important component in simulation applications which are based on virtual geographic information system (VGIS). In this paper, an effective collision detection algorithm for multiple objects in VGIS, VGIS-COLLIDE, is presented. The algorithm firstly integrates existing quadtree, which is the global hierarchical structure of VGIS, with axis-aligned bounding box of object to perform the broad-phase of collision detection. After that, exact collision detection between two objects which have passed the broad-phase of collision detection is performed. The algorithm makes no assumption about input primitives or object's motion and is directly applicable to all triangulated models. It can be applicable to both rigid and deformable objects without preprocessing. The performance of the algorithm has been demonstrated in several environments consisting of a high number of objects with hundreds of thousands of triangles.